Method and system for managing resource allocation in non-uniform resource access computer systems

Abstract

A method and system of managing resource allocation in a non-uniform resource access computer system is disclosed. A method comprises determining access costs between resources in a computer system having non-uniform access costs between the resources. The method also includes constructing a hierarchical data structure comprising the access costs. The hierarchical data structure is traversed to manage a set of the resources.

Description

BACKGROUND OF THE INVENTION FIELD OF THE INVENTION [0001]The present invention relates to the field of non-uniform resource access computer systems. Specifically, embodiments of the present invention relate to methods and systems providing a hierarchical data structure describing access costs to resources in order to facilitate managing the resources.BACKGROUND ART [0002]Uniform Memory Access (UMA) computer systems have the characteristic of a processor, for example, CPU (Central Processing Unit), having essentially the same access time to all memory resources. There may be multiple CPUs in such a system, each with the characteristic of little or no performance difference for one memory resource over another based on access cost. In such a system, there is little or no benefit in assigning memory resources to a given CPU based on expected access time. In contrast, for Non-Uniform Memory Access (NUMA) computer systems, the cost for a CPU to access memory can vary significantly for di...

AI Technical Summary

Benefits of technology

[0006]The present invention provides methods and systems of facilitating resource management in a non-uniform resource access computer system. Embodiments of the present invention allow a resource allocator to easily decide between a resource with the lowest access cost, a next lowest access cost, etc. Embodiments of the present invention may be transparent to application programs, and hence do not require modification to the application programs. However, embodiments of the present invention may also provide access cost information to applications to allow observability and the opportunity to use the information to customize its resource allocation as it pleases. Embodiments of the present invention facilitate efficient allocation of resources in a non-uniform resource access computer system. The present invention provides these advantages and others not specifically mentioned above but described in the sections to follow.

[0007]Embodiments of the present invention provide an abstraction for the operating system and application programs to use for improving performance on NUMA machines. An access cost group is used to represent a group of resources that share a common access cost value. For example, an access cost group may contain a group of CPUs and a memory resource, wherein each of the CPUs has the same access cost value to the memory resource, or vice versa. An access cost function defines how the access cost value is calculated and may be defined by any suitable combination of parameters. Access cost groups are arranged in a hierarchical data structure such that children contain resources that are closer together than the resources contained in the child's parent(s). The hierarchical data structure makes it easy for the operating system and application programs to determine what resources are close to each other without knowing the details of the hardware. The hierarchical data structure makes it possible to determine not only the closest resource, but also the next closest resource and each successive closest resource.

Problems solved by technology

However, the hardware configuration information that is typically collected does not easily facilitate the allocation of resources.

It is difficult to base a resource allocation decision based on such hardware configuration information.

Moreover, modifying the application program to take advantage of the hardware configuration information makes the application program less portable.

While this technique is simple, it results in slower execution than would be possible with a more intelligent allocation of CPU and memory resources.

While the above discusses problems with allocating CPU and memory resources, allocating resources other than CPU and memory resources also presents problems for conventional techniques.

Therefore, a problem with conventional methods of allocating resources in non-uniform resource access computer systems is the difficulty in basing a resource allocation decision on hardware configuration information.

Another problem with conventional methods of allocating resources in non-uniform resource access computer systems is that application programs need to be modified to allow them to take advantage of hardware configuration information.

Another problem is the inefficient allocation of resources when resources are allocated without regard to access costs.

Images